Neutron-star mergers and the r-process

How is gold made? Answer 2: the fast route

The s-process builds heavy nuclei by waiting. The r-process builds them by not waiting at all. The s-process explains roughly half of the cosmic abundances of elements heavier than iron, but its signature is carried mostly by nuclei close to the valley of beta stability, along a path that follows the valley itself and stops at lead. The other half of the heavy elements — the actinides $\mathrm{Th}$ and $\mathrm{U}$, the noble metals $\mathrm{Pt}$ and $\mathrm{Au}$, rare earths such as $\mathrm{Eu}$, $\mathrm{Tb}$, $\mathrm{Dy}$ and $\mathrm{Yb}$, and the markedly neutron-rich isotopes at intermediate masses — requires a completely different mechanism, in which the neutron flux is many trillion times more intense. When $n_n$ reaches $10^{20}$-$10^{25}$ cm$^{-3}$ and the temperature rises to $T \sim 10^{9}$ K, the mean time between captures falls to $\tau_n \sim 10^{-4}$ s, while beta-decay lifetimes remain closer to tenths of seconds or seconds. A nucleus has no time to decay between one capture and the next: the chain of absorptions proceeds far from the valley of stability, all the way to the neutron drip line, where adding one more neutron is no longer bound. This is the r-process (rapid neutron capture), proposed in quantitative form by Burbidge, Burbidge, Fowler and Hoyle in 1957 and independently by Cameron in the same year [Cameron 1957], and for more than half a century a nuclear mechanism in search of an astrophysical site.

The answer arrived on 17 August 2017, when the LIGO-Virgo network detected the gravitational waves of a coalescing pair of neutron stars and, in the following hours, dozens of telescopes observed its electromagnetic counterpart carrying the spectroscopic signature of freshly synthesized lanthanides and actinides. The story of that confirmation — and the physics that made it possible — occupies the second half of this chapter. First, however, we need the kinetic picture of the process, its appetite for nuclear data, and the map of candidate astrophysical sites that competed for the role for fifty years.

The kinetic picture

The formal condition for the r-process is $\tau_n \ll \tau_\beta$ for all nuclei along the path, including those at the extreme edge of neutron stability. The resulting geometry on the chart of nuclides is radically different from that of the s-process: instead of zigzagging along the valley of stability, the r-process flow pushes toward the drip line, runs along isotones $N = \mathrm{const}$ at the magic numbers $N = 50, 82, 126$, and piles up at “waiting points” — nuclei where the equilibrium between $(n,\gamma)$ capture and $(\gamma,n)$ photodisintegration at high temperature temporarily blocks the flow, until beta decay unlocks the next nuclide along the isotone.

In $(n,\gamma) \rightleftharpoons (\gamma,n)$ equilibrium, the relative abundance of isotopes along an isotopic chain obeys a Saha-like relation,

$$\frac{N(Z, A+1)}{N(Z, A)} = n_n \, \left(\frac{2\pi m_u kT}{h^2}\right)^{-3/2} \frac{G(Z, A+1)}{2\,G(Z, A)} \exp\left(\frac{S_n(Z, A+1)}{kT}\right),$$

where $S_n$ is the neutron separation energy and $G$ the nuclear partition function. Along each isotopic chain there is a critical nuclide for which $N(Z, A+1)/N(Z, A) \sim 1$: this is the waiting point. Its location depends on $n_n$, $T$ and $S_n$ — quantities to which the picture is sensitive, and which constrain the conditions of the site once the final abundance pattern is observed.

When the neutron density drops below the threshold that sustains the equilibrium — the freeze-out, on a timescale of a second — the waiting points are no longer maintained, and the nuclides of the chain beta-decay in cascade toward the valley of stability, crossing 5-15 elements and populating the stable nuclei we observe today. The consequence on the cosmic abundance curve is a sequence of three prominent r-process peaks at $A \approx 80, 130, 195$, shifted to lower $A$ with respect to the corresponding s-process peaks at $A \approx 88, 138, 208$: the r-process peak at the magic number $N = 82$ forms from waiting points around $Z \approx 45$-$48$ ($A \approx 127$-$130$), and since the beta cascade conserves mass number (apart from modest delayed-neutron emission), the observed peak remains at $A \approx 130$ — below the s-process peak at $A \approx 138$, which instead forms at the valley of stability, where $N = 82$ falls on barium. The same reasoning applies to the other peaks. The separation between s and r peaks in the solar abundance curve is the most immediate diagnostic signature of the two processes.

For the heaviest nuclei ($Z \gtrsim 95$), the r-process path meets a new threshold: spontaneous fission and neutron-induced fission become competitive with capture, and the flow stops. Fission fragments re-inject nuclei into the range $A \sim 130$-$160$ — the so-called fission cycling, active only for very high neutron exposures ($\tau \gtrsim 1$ mb$^{-1}$), which plays a crucial role in establishing the universality of the observed r-process pattern (see the section on abundance patterns, later in this chapter).

Nuclear data

The quantitative calculation of an r-process path requires masses and Q-values for roughly $10^{4}$ nuclides, beta-decay widths and beta-delayed neutron emission probabilities for as many, $(n,\gamma)$ cross sections at $T \sim 10^{9}$ K, and spontaneous and induced fission rates for the heaviest nuclei. Almost all of these quantities are extrapolations from theoretical models: the relevant nuclei lie at or beyond the limit of experimental production even at the most modern facilities (FRIB at MSU, RIBF at RIKEN, FAIR at Darmstadt). The reference mass formulas are FRDM (Möller-Nix-Kratz), HFB-27 (Goriely) and Duflo-Zuker, with differences up to a few MeV for the most exotic nuclei and significant impact on the calculated final pattern. Beta rates are computed in the QRPA approximation, with typical uncertainties of a factor 2-3. The $(n,\gamma)$ cross sections are computed in the Hauser-Feshbach approximation, sensitive to level densities and to the $\gamma$-strength function — uncertainties of a factor 5-10 for the most distant nuclei.

The most widely used r-process nucleosynthesis codes are SkyNet (Lippuner-Roberts, open-source), the JINA r-process code (FSU/MSU), WinNet (GSI-Basel) and PRISM (LANL). The canonical, up-to-date review of the field is Cowan, Sneden, Lawler and collaborators (2021) [Cowan et al. 2021] in Reviews of Modern Physics, today the standard reference.

Astrophysical sites: the fifty-year problem

For more than half a century, from the publication of B²FH to 2017, the great open question of nucleosynthesis was: where does the r-process happen? The required conditions — $n_n \sim 10^{24}$ cm$^{-3}$, $T \sim 10^{9}$ K, strongly neutron-rich matter with $Y_e = n_p/(n_p + n_n) \lesssim 0{.}25$ — are extreme and occur in no standard stellar site. Proposals followed one another for decades: core-collapse supernovae (in particular the neutrino-driven wind from the proto-neutron star), exotic supernovae with magneto-rotational jets, mergers of neutron stars in compact binary systems. None had been confirmed observationally, and each had significant quantitative difficulties.

The map of candidate sites is summarized in the following table:

The picture was resolved in large part — though not yet completely — by the event of 17 August 2017, told in the next section: the simultaneous detection of gravitational waves, short gamma rays, optical/UV/NIR, X-rays and radio showed in detail the behavior of a binary neutron-star merger (NSM), and its kilonova carries the spectroscopic signature of an active r-process site. It remains open whether NSM are the only sites of the r-process, or whether other sites — rarer but more productive per event, or more frequent but less productive — contribute significantly to the Galactic budget: the quantitative comparison between candidate sites and the overall Galactic budget is taken up in the final part of the chapter.

The event of 17 August 2017

On 17 August 2017 at 12:41:04 UTC, the advanced LIGO detectors at Hanford and Livingston and the Virgo interferometer at Cascina recorded, for about one hundred seconds, a gravitational-wave signal of rising frequency — the classic chirp of two coalescing masses — from a sky direction localized to a region of about 30 square degrees south of the celestial equator [Collaboration & Collaboration 2017] . The waveform was unmistakably that of a binary neutron-star merger (BNS), with individual masses between $1{.}1$ and $1{.}6\,M_\odot$ (at 90% confidence) and a luminosity distance of $40 \pm 8$ Mpc from the standard relation between chirp amplitude and frequency. Almost two seconds after coalescence, the Fermi/GBM and INTEGRAL satellites independently detected a short gamma-ray burst from the same direction (GRB 170817A, $T_{90} = 2{.}0$ s, isotropic luminosity $L_\gamma \approx 1{.}6 \times 10^{47}$ erg/s — anomalously under-luminous compared with typical short GRBs, and later interpreted as an off-axis view of a structured relativistic jet). In the following hours, dozens of telescopes pointed at the localization region: eleven hours after the merger, the Swope telescope at Las Campanas identified a new optical source in NGC 4993, immediately catalogued as AT2017gfo (alias SSS17a, DLT17ck). Its photometric and spectroscopic evolution over the following days and weeks matched in detail the theoretical predictions for a kilonova developed by Li, Paczyński, Metzger, Kasen, Barnes and collaborators over the previous twenty years — the optical-infrared glow expected from ejecta heated by the radioactive decay of freshly produced r-process elements.

The detailed study of AT2017gfo provided the first direct confirmation — hypothesized since the 1970s by Lattimer-Schramm and Eichler-Livio-Piran-Schramm — that neutron-star mergers are indeed r-process sites, and produce the heaviest elements of the universe in significant quantities: the total mass of r-elements produced in GW170817 has been estimated at $0{.}02$-$0{.}05\,M_\odot$, enough to contain a quantity of gold equal to about one to two hundred Earth masses. The event formally opened the era of multimessenger astronomy, in which a single astrophysical event is observed simultaneously in the gravitational channel, the electromagnetic channel (from high-energy gamma rays to radio) and — in principle, for Galactic events — the neutrino channel.

The multiband observational coverage of GW170817/AT2017gfo was of historic scope. In gravitational waves, the LIGO/Virgo network constrained the masses, the distance and — through the tidal deformability of the late signal — the compactness of the two stars. In gamma rays, GRB 170817A was the first short GRB unambiguously associated with a BNS event. In the optical, UV and near-infrared, AT2017gfo was followed by Swope, DECam, MASTER, REM and some seventy other telescopes in the first 24 hours, with high-resolution spectroscopy from XSHOOTER at the VLT, FLAMINGOS-2 at Gemini South, and many other instruments over the following months. In X-rays, Chandra and XMM-Newton detected emission about 9 days post-merger, which grew and then faded over months to years, in agreement with the expected evolution of an off-axis relativistic-jet afterglow. In radio, VLA, GMRT and ATCA detected emission about 16 days post-merger, with temporal and spectral evolution consistent with a structured jet. The spectroscopic identification of strontium in AT2017gfo by Watson and collaborators (2019) [Watson et al. 2019] , through a feature around 800 nm consistent with $\mathrm{Sr}^{+}$, provided the first direct signature of a single light r-process element ($A \approx 90$); the later identification of tellurium (Hotokezaka et al. 2023) [Hotokezaka et al. 2023] , through an emission line in the mid-infrared at $\sim 2{.}1$ µm in late-time spectra, extended the identification to elements of the second r-process peak.

The observational statistics of NSM after 2017 are still limited. GW190425 (LIGO/Virgo, run O3a) is the second confirmed BNS event, with an anomalously high total mass ($\sim 3{.}4\,M_\odot$, exceeding the largest BNS mass observed in Galactic binary pulsars): no optical counterpart was identified, both because of the very large localization region and because of the greater distance. GW230529 in O4 is classified as a mass-gap merger (one component in $2{.}5$-$5\,M_\odot$, probably NS-BH). A dozen additional BNS candidates in runs O3 and O4 were recorded at reduced sensitivity due to intermittent unavailability of the Virgo component. The local rate of BNS mergers inferred from these detections is $\sim 100$-$1500$ Gpc$^{-3}$ yr$^{-1}$ (90% C.L.), corresponding to $\sim 10^{-5}$-$10^{-4}$ per year in the Milky Way, consistent with independent estimates from Galactic binary pulsars. For NS-BH the inferred rate is about an order of magnitude lower, with larger uncertainties.

Phases of the coalescence

A binary neutron-star merger proceeds through four temporally distinct phases, each with its own dominant physics and its own observables. The inspiral phase covers the entire dynamical history of the system up to contact: two NS in a binary orbit lose energy to gravitational radiation at the rate predicted by Einstein’s quadrupole formula, and the orbit progressively shrinks. In the last hundred seconds before merger (for typical stellar masses), the orbital frequency sweeps from $\sim 1$ Hz to $\sim 1$ kHz and the gravitational-wave amplitude grows as $f^{2/3}$. The chirp — a GW signal of rising frequency and amplitude — is the signature detected by ground-based interferometers, and from its shape one extracts the individual masses, the total mass and (in the last milliseconds) the tidal deformability that encodes the equation of state of neutron matter.

The merger phase proper lasts a few milliseconds. The two NS come into contact at GW frequency $\sim 1$ kHz and separation $r \sim 30$ km — comparable to their individual radii. The dynamics is dominated by asymmetric tidal torques and by hydrodynamic shocks at the contact interface. The immediate outcome is a central object that, depending on the total mass relative to the maximum TOV mass of the EOS, can be a hypermassive neutron star (HMNS) transiently supported by differential rotation, or collapse directly to a black hole within a few milliseconds. At the same time, a small fraction of the total mass is expelled into the surrounding medium as dynamical ejecta ($M \sim 10^{-3}$-$10^{-2}\,M_\odot$): the tidal component is expelled along the equatorial plane by tidal torques acting on the outermost layers, is very neutron-rich ($Y_e \sim 0{.}05$-$0{.}15$) and typically cold; the shock component is expelled along the polar axis by compression at the interface, is less neutron-rich ($Y_e \sim 0{.}15$-$0{.}25$) and hotter.

The post-merger phase covers the following seconds and includes the formation and evolution of the residual accretion disk around the central remnant (HMNS or BH). The disk has typical mass $\sim 0{.}1\,M_\odot$, temperatures $\sim 10^{10}$ K, and densities that justify cooling by intense neutrino emission. Over $\sim 1$-$10$ s, the combined action of magneto-hydrodynamic viscosity, recombination of nucleons into $\alpha$ particles, and neutrino reabsorption in the inner layers drives a disk wind ejecta of mass $\sim 10^{-2}$-$10^{-1}\,M_\odot$ — the most massive component of the total ejecta — with proton fraction $Y_e \sim 0{.}2$-$0{.}4$ (higher than the dynamical ejecta because partially neutrino-processed by the $\nu_e$ flux). If the central remnant is a BH (or becomes one quickly), a collimated relativistic jet can emerge with Lorentz factor $\Gamma \sim 100$-$1000$ and isotropic-equivalent energy $\sim 10^{49}$-$10^{50}$ erg: it is the origin of the observed short gamma-ray burst. The conditions of the different ejecta components are summarized in the following table:

The kilonova phase covers the following days and weeks, and is dominated by the radiative cooling of the ejecta heated by the $\beta^{-}$ decay of the freshly produced r-process elements. At $t \sim 1$ day post-merger, the ejecta occupy a volume $\sim 10^{42}$ cm$^{3}$, expand at $v \sim 0{.}1$-$0{.}3\,c$, and have density $\sim 10^{-13}$ g/cm$^{3}$. The light curve has two spectrally distinct components. The blue kilonova is produced by ejecta with $Y_e > 0{.}25$, dominated by light r-process elements ($A \lesssim 130$) whose opacity is low: UV/blue peak at $\sim 1$ day with $L \sim 10^{42}$ erg/s, declining within a few days. The red kilonova is produced by ejecta with $Y_e < 0{.}25$, dominated by lanthanides (with a high line-density $4f^{n}$ ground configuration) and actinides that produce very high optical and infrared opacity: IR peak at $\sim 1$ week with $L \sim 10^{41}$ erg/s, lasting weeks. The superposition of the two components in time and wavelength, observed in AT2017gfo, is the direct signature of ejecta with different $Y_e$ compositions — an indirect but strong test of the hydrodynamic structure of the merger.

GR-MHD simulations of BNS mergers (Whisky in Trento, SACRA in Kyoto, Kentucky/IllinoisGRMHD, FORNAX at Princeton, ALCAR at Garching) now reach resolutions of $\sim 100$ m in the core during the merger and follow the post-merger evolution to about 100 ms (some recent simulations to $\sim 1$ s with reduced magnetic-field treatment). The resulting constraints on the NS equation of state, combined with independent measurements of massive pulsars and NICER radii, are progressively tighter. The maximum TOV mass observed in pulsars is $M_{\max} > 1{.}97\,M_\odot$ (PSR J0740+6620, measured via Shapiro delay), with recent measurements pushing the limit toward $\sim 2{.}1\,M_\odot$. The canonical radius of a $1{.}4\,M_\odot$ NS is $R_{1{.}4} = 11{.}9 \pm 1{.}0$ km from the combination of NICER and the tidal deformability of GW170817 ($\Lambda_{1{.}4} = 190^{+390}_{-120}$ at 90% C.L.). The EOS currently consistent with all constraints are APR, SLy, DD2 and variants, possibly with a phase transition to deconfined quarks at the highest central densities.

R-process nucleosynthesis in NSM

The matter expelled by a neutron-star merger is extremely rich in neutrons: the proton fraction $Y_e = n_p/(n_p + n_n)$ is typically $0{.}1$-$0{.}3$, against the $0{.}5$ of ordinary interstellar gas. For the tidal dynamical ejecta, $Y_e$ can drop to $0{.}05$-$0{.}1$, values that directly reflect the composition of the outermost layers of the original NS — nearly pure neutron matter with a small fraction of degenerate electrons. This is exactly the regime of the strong r-process described in the kinetic picture at the opening of this chapter: the seed nuclei (starting from $\alpha$ particles and the few heavy nuclei pre-existing in the NS crust), heated to several billion kelvin in the first milliseconds post-merger, capture neutrons rapidly — with mean time between captures $\tau_n \sim 10^{-4}$ s — reaching the neutron drip line, proceeding along the magic-number isotones, and arriving at the actinides; at freeze-out, the net flow ceases and the whole composition beta-decays in cascade to the valley of stability, populating the final nuclides we observe.

The total r-process mass produced in GW170817 has been estimated from combined modeling of the kilonova light curve at $\sim 0{.}02$-$0{.}05\,M_\odot$, of which $\sim 0{.}01\,M_\odot$ in the strong dynamical ejecta and the rest in the weak disk-wind ejecta. The total mass of lanthanides produced is $\sim 10^{-3}\,M_\odot$, and the gold $\sim 100$-$200\,M_\oplus$. In the very strong regime of the dynamical ejecta, fission cycling is also active — the regeneration of seed nuclei by fission of the heaviest nuclei — which “washes” the final pattern of its initial conditions; the mechanism and its consequences for the universality of the pattern are discussed in the next section.

Kilonova models coupled to detailed r-process networks (Kasen [Kasen et al. 2017] , Tanaka, Wanajo, Barnes) compute the opacity for each lanthanide species integrated over all available electronic transitions — a non-trivial calculation, because the $4f^{n}$ ground configuration of the lanthanide ions $\mathrm{Ln}^{2+,3+}$ produces $> 10^{6}$ atomic transitions in the visible-IR — and produce synthetic spectra via diffusion approximation or detailed radiative transfer (TARDIS, SUMO, JEDD). The spectral identification of Sr, Te and La in AT2017gfo was possible only thanks to these quantitative line-opacity calculations. The residual uncertainties of the models depend on the equation of state of neutron matter (which conditions the fate of the central remnant and hence the blue/red ratio of the kilonova), on the r-process opacities themselves (computed from atomic-structure models for lanthanides and actinides 3+, still with factor 2-3 uncertainties), on the relative extent of the weak and strong r-process (constrained by the shape of the lanthanide peak), and on the mass ratio $q = M_2/M_1$ of the binary system (which conditions the mass of dynamical ejecta).

Abundance patterns and the r-process signature

The abundances of r-process elements in the Solar System are obtained by subtraction: from the total solar abundances one subtracts the calculated contributions of the s-process (based on calibrated AGB models, see chapter 4) and of the p-process (chapter 5), and the residual is attributed to the r-process. The resulting curve shows a characteristic structure: the peak at $A \approx 80$ (Se, Br, Kr), originating from the weak r-process or from an intermediate component with $Y_e \sim 0{.}3$-$0{.}4$; the peak at $A \approx 130$ (Te, I, Xe, Ba), originating from the main r-process at the $N = 82$ shell closure; the “rare earth peak” at $A \approx 165$ (Tb, Dy, Ho), originating from nuclear deformation in the rare-earth region and particularly sensitive to nuclear masses; the peak at $A \approx 195$ (Os, Ir, Pt), at the $N = 126$ shell closure; and the actinide regime (Th, U), produced at the end of the path and partially erased by fission.

One of the most surprising discoveries — and still today one of the most discussed — of metal-poor stellar spectroscopy is the universality of the main r-process pattern in EMP stars. The abundance curve for $Z = 56$-$80$ (from Ba to Hg) in r-II stars — the strongly enriched ones, with $[\mathrm{Eu/Fe}] > 1$ — is scalable to the solar curve within $\pm 0{.}1$ dex for the great majority of nuclides. Stars living in different cosmological epochs (from $z \sim 6$-$10$ to today), in different chemical environments (Galactic halo, dwarf spheroidal galaxies, Solar System), carry the same nuclear signature. The implication is that the main site of the main r-process always produces the same pattern of relative abundances, regardless of the details of the initial conditions — a property that demands a robust physical explanation.

The most convincing explanation, currently under discussion, is fission cycling: in a very strong r-process ($\tau > 1$ mb$^{-1}$), the capture chain reaches fissionable nuclei in the region $A \sim 250$-$280$ (the region of Cf, Es, Fm, with characteristic spontaneous-fission times of order seconds at freeze-out temperatures), which fission spontaneously or by neutron capture, re-injecting two fragments with an approximately Gaussian mass distribution into the range $A \sim 130$-$160$, where they serve as seeds for a new cycle. After a few fission cycles, the final composition “forgets” the initial conditions and converges to an attractor determined solely by the nuclear structure along the path — the r peaks at the magic numbers and the rare-earth peak in the deformation regime. An alternative explanation (Aoki, Honda, Mathews) attributes the universality to the $(n,\gamma) \rightleftharpoons (\gamma,n)$ equilibrium during freeze-out, which locks the abundances at the magic-number waiting points before the beta cascade: in this scenario, universality is a local property of the freeze-out rather than a global property of the path, and should hold even for moderate exposures.

For $Z < 56$ (below barium) and $Z > 80$ (above mercury), the pattern is not universal: individual EMP stars show significant dispersion, and in some (the so-called actinide boost stars such as CS 31082-001) the $[\mathrm{Th/Eu}]$ ratio is significantly overabundant with respect to the solar value. The commonly accepted conclusion is that the main r-process (responsible for the region $A = 130$-$195$) is robust and universal, while the weak r-process (above the first peak) and the actinide regime vary from site to site, reflecting different conditions of $Y_e$ and exposure duration. This is consistent with a picture in which NSM have multiple ejecta components (dynamical, disk wind, neutrino-driven) with different $Y_e$, and in which the relative mixture varies from event to event. Self-consistent 3D GR-MHD simulations of the post-merger disk, which are beginning to become available (Fernández-Foucart-Metzger, Just-Bauswein), promise to clarify the $Y_e$ distribution in the ejecta and to provide a quantitative picture of the weak component.

The r-II stars are the “refrigerator of history” of r-process nucleosynthesis: they freeze the signature of a single nearby r-process event, because the matter they are made of saw only that event (or a few correlated events) before their formation. Their abundance in the Galactic halo — about $5\%$ of EMP stars, counting together the r-I with $0{.}3 < [\mathrm{Eu/Fe}] < 1$ and the r-II with $[\mathrm{Eu/Fe}] > 1$ — is consistent with the expected NSM rate in the primordial population, taking into account the delay time from the progenitor SN II to the coalescence of the NSM (typically $10^{7}$-$10^{9}$ years). The most complete methodological review of these systems and of the constraints they impose on r-process sites is Thielemann, Eichler, Panov and Wehmeyer (2017) [Thielemann et al. 2017] .

Contribution to the Galactic budget

After GW170817 it was clear that NSM produce r-process material. What remained open — and to a large extent remains open today — is how much of the overall Galactic budget of r-process elements comes from NSM and how much from other sites. Current answers, based on several converging methods of analysis, place the NSM contribution at $30$-$70\%$ of the total Galactic budget, with the rest plausibly attributed to the exotic core-collapse supernovae already met in the map of candidate sites (collapsars, MHD-jet SNe, magnetar SNe).

The quantitative constraints on the NSM contribution rest on three key numbers: the local rate of BNS mergers deduced by LIGO ($\sim 300^{+700}_{-200}$ Gpc$^{-3}$ yr$^{-1}$ in O3, at 90% C.L., with progressively smaller uncertainties in the coming runs); the Galactic rate estimated by scaling ($\sim 10^{-5}$-$10^{-4}$ per year, consistent with counts of Galactic binary pulsars); and the mean r-process mass produced per merger ($\sim 0{.}02$-$0{.}1\,M_\odot$, modulated by mass ratio, EOS, and the fate of the central remnant). Combined, these numbers yield a cumulative mass of r-elements from NSM over a Hubble time of order $10^{3}$-$10^{4}\,M_\odot$ — sufficient or sub-optimal with respect to the total Galactic budget of $\sim 10^{4}\,M_\odot$, depending on the exact assumptions.

Independent observational constraints come from three directions. The dispersion of $[\mathrm{Eu/Fe}]$ vs $[\mathrm{Fe/H}]$ in halo EMP stars — about $\pm 1$ dex at $[\mathrm{Fe/H}] < -2$ — is naturally explained by rare and productive r events, consistent with the NSM-dominant picture. The $^{244}\mathrm{Pu}$ ($\tau_{1/2} = 8 \times 10^{7}$ yr) detected in deep-sea sediments at levels $\sim 10^{-15}$ atoms/g (Wallner et al. 2015) [Wallner et al. 2015] implies a local r source within a few hundred million years — compatible with the expected NSM rate within the 100 pc surrounding the Sun. The stars of the ultra-faint dwarf galaxy Reticulum II show an extreme excess of $[\mathrm{Eu/Fe}]$ compared with other comparable dwarfs (Ji et al. 2016) [Ji et al. 2016] , interpreted as the signature of a single ancestral r-process event that enriched the entire galaxy — an event whose mass and frequency are consistent with an NSM.

The quantitative debate remains alive. Côté et al. (2018) [Côté et al. 2018] argue that the delay time distribution of NSM (with median value $\sim 1$ Gyr) is too long to explain the presence of r-elements in the oldest halo stars with $[\mathrm{Fe/H}] < -3$, and propose that exotic supernovae are the complementary “fast” source needed to seed the primordial r enrichment. Other analyses (Hotokezaka et al. 2018, Wehmeyer et al. 2019, Beniamini et al. 2018) maintain that NSM with short delay times (NSM from massive Pop II or late-Pop III progenitors) are sufficient, and that the contribution from alternative sites is subdominant. The final resolution depends on three converging fronts: precise characterization of the NSM delay-time distribution (observable from future GW surveys combined with host-galaxy properties); constraints on the fraction of “exotic” core-collapse SNe (from long GRBs and orphan GRB afterglows); and constraints on the mean yield per NSM (from self-consistent 3D hydrodynamic models and from spectroscopy of future kilonovae). The methodological review of the field is today Cowan, Sneden, Lawler et al. (2021) [Cowan et al. 2021] , complemented by the r-process framework of Thielemann et al. (2017) [Thielemann et al. 2017] .

Prospects: the multimessenger decade

Multimessenger astronomy is just beginning. With the next gravitational-wave detectors and the release of new-generation multiband datasets, the number of observed and characterized NSM will grow by three orders of magnitude over the next ten to fifteen years. Each event will be accompanied — when the localization region and the distance allow it — by detailed spectrophotometric follow-up of its kilonova. In parallel, the new nuclear-physics laboratories will directly measure the properties of nuclei far from stability that today are only theoretically extrapolated, significantly reducing the uncertainties of r-process networks.

The instrumental roadmap over 10-15 years unfolds on several converging fronts, summarized in the following table:

The key questions one hopes to answer quantitatively within this horizon are at least four. The first is the precise fraction of the Galactic r-process budget from NSM vs other sites, constrained by the combination of NSM rate, delay-time distribution, and the statistics of r-process-enhanced stars. The second is the r-process abundance pattern as a function of merger type (standard BNS vs NS-BH vs BNS with asymmetric masses): kilonovae with different mean $Y_e$ compositions should produce slightly different patterns, and a statistical sample of tens to hundreds of spectrally resolved events will test this prediction. The third is the NS equation of state above $\rho_{\mathrm{nuc}}$, constrained by the tidal deformability of GW events and by NICER radii, in convergence with massive-pulsar constraints. The fourth is the identification of specific atomic lines of heavy r-elements in kilonova spectra: beyond the already identified Sr, Te and La, the goal is to isolate signatures of Pt, Au, Ce, Nd and U, which would require high-resolution spectroscopy of sufficiently nearby events. On a complementary front, ELT, GMT and TMT will measure with precision the spectra of hundreds of r-I and r-II stars in the halo and in satellite dwarf galaxies, and in some cases the isotopic abundances of Eu, Ba and Nd — a direct constraint on the isotopic ratios produced by the site and a complementary probe of the freeze-out conditions.

The observational programs currently feeding these goals include ENGRAVE (European multi-Messenger Observations of GW sources), the European consortium for rapid kilonova follow-up with priority access to VLT, NTT, GTC and other telescopes of the ESO community; GROWTH (Global Relay of Observatories Watching Transients Happen), the Caltech-coordinated network with access to Palomar, Keck, DECam and others; MMA-Argo, a dedicated marine-sediment sampling program for the detection of $^{244}\mathrm{Pu}$ and $^{60}\mathrm{Fe}$ as tracers of recent r-process events; and the R-Process Alliance, the international consortium for spectroscopy of r-enhanced stars, which has so far catalogued more than 200 r-I and r-II stars in the Galactic halo. The observational effort is coordinated with the simulation campaigns (NICER + EOS, 3D GR-MHD post-merger, kilonova radiative transfer) and with the laboratory measurement campaigns (FRIB FY24-28 scientific program, ELI-NP photodisintegration, with RIBF at RIKEN and FAIR at Darmstadt completing the map of drip-line nuclei). The combination promises to close — not structurally, but quantitatively — the main open ambiguities of r-process nucleosynthesis within the next decade.

GW170817 transformed the r-process from a nuclear mechanism in search of a site into an astrophysical phenomenon at least partially observed in real time, and formally opened the era of multimessenger astronomy. The r-process also closes the picture of heavy-element nucleosynthesis by neutron capture: together with the s-process of chapter 4 it covers practically all the trans-iron abundances of the universe, while the small fraction of proton-rich nuclei that neither process produces is covered by the p-process treated in chapter 5. What remains to be told is the bridge between individual nucleosynthesis events — explosive (chapter 5), compact (this chapter) or quiescent — and the statistical picture of the cosmic abundances observed in the Sun, in the interstellar medium and in stars: the chemical abundance curve is the starting datum and the ultimate test bench of the whole discipline, and the next chapter is devoted to it.